What is the best way to ensure a model does not provide non-existent URLs in its responses? I've been struggling with this for a while now...
#non-existent URLs provided
uneven gulch
void mothBOT
Hi @uneven gulch. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
Based on the information provided in the knowledge sources, there are a few strategies you can use to ensure a model does not provide non-existent URLs in its responses:
-
Avoid having the model construct full URLs on the fly. Instead, you should pass the full URL back in an Action response. This can help ensure that the URL is valid and exists, as you can verify it before passing it to the model.
-
Avoid having the model fill in URL details. Similar to the first point, you should pass in the full URL and all UTM parameters. This can help prevent the model from generating incorrect or non-existent URLs.
-
If the URLs are coming from your existing dataset, ensure that the dataset only contains valid and existing URLs. If the URLs are not from your dataset, you may need to rely on some local function in your system to perform web searches or validate the URLs.
Remember, a large language model is mostly meant for natural language processing of some kind and it's not meant as a (direct) search engine. So, it might not be able to accurately generate or validate URLs on its own.
Sources:
- Issue with Generated but non-clickable Links in Assistant Output (https://community.openai.com/t/567342)
- Is there a way to API responses to providing only URLs? (https://community.openai.com/t/651209)